Identification Method of Ancient Ceramics Revision

摘要

Along with the promotion of economics and cultural, more and more people adore collecting some arts, for example, ancient or modern ceramic and so on. Ancient ceramic has significant research and collectible value of its historical and humanistic value. It only presents some subtle changes on shape and rare surfaces in comparison with others. The key problem of the identification of ancient ceramics is a conflict between the high-performance requests and the low accuracy and precise of algorithm. The paper proposes a novel identification method of ancient ceramics based on VGG16-GSSVM with few-shot. It uses the pre-trained VGG16 to extract the shape features of ancient ceramics, and trains a SVM model to be the classifier of VGG16. It applies Grid Search (GS: Grid Search) algorithm to determine its optimal parameters of SVM. Finally, the paper uses VGG16-GSSVM to identify pottery. The experiment shows that VGG16-GSSVM model has a higher accuracy and efficiency than any of CNN and CNN-SVM model while the ratio of the training to test is 1:3.